Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/torvalds/linux-2.6
[linux-2.6] / net / sched / sch_prio.c
1 /*
2  * net/sched/sch_prio.c Simple 3-band priority "scheduler".
3  *
4  *              This program is free software; you can redistribute it and/or
5  *              modify it under the terms of the GNU General Public License
6  *              as published by the Free Software Foundation; either version
7  *              2 of the License, or (at your option) any later version.
8  *
9  * Authors:     Alexey Kuznetsov, <kuznet@ms2.inr.ac.ru>
10  * Fixes:       19990609: J Hadi Salim <hadi@nortelnetworks.com>:
11  *              Init --  EINVAL when opt undefined
12  */
13
14 #include <linux/module.h>
15 #include <linux/types.h>
16 #include <linux/kernel.h>
17 #include <linux/string.h>
18 #include <linux/errno.h>
19 #include <linux/skbuff.h>
20 #include <net/netlink.h>
21 #include <net/pkt_sched.h>
22
23
24 struct prio_sched_data
25 {
26         int bands;
27         struct tcf_proto *filter_list;
28         u8  prio2band[TC_PRIO_MAX+1];
29         struct Qdisc *queues[TCQ_PRIO_BANDS];
30 };
31
32
33 static struct Qdisc *
34 prio_classify(struct sk_buff *skb, struct Qdisc *sch, int *qerr)
35 {
36         struct prio_sched_data *q = qdisc_priv(sch);
37         u32 band = skb->priority;
38         struct tcf_result res;
39         int err;
40
41         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
42         if (TC_H_MAJ(skb->priority) != sch->handle) {
43                 err = tc_classify(skb, q->filter_list, &res);
44 #ifdef CONFIG_NET_CLS_ACT
45                 switch (err) {
46                 case TC_ACT_STOLEN:
47                 case TC_ACT_QUEUED:
48                         *qerr = NET_XMIT_SUCCESS | __NET_XMIT_STOLEN;
49                 case TC_ACT_SHOT:
50                         return NULL;
51                 }
52 #endif
53                 if (!q->filter_list || err < 0) {
54                         if (TC_H_MAJ(band))
55                                 band = 0;
56                         return q->queues[q->prio2band[band&TC_PRIO_MAX]];
57                 }
58                 band = res.classid;
59         }
60         band = TC_H_MIN(band) - 1;
61         if (band >= q->bands)
62                 return q->queues[q->prio2band[0]];
63
64         return q->queues[band];
65 }
66
67 static int
68 prio_enqueue(struct sk_buff *skb, struct Qdisc *sch)
69 {
70         struct Qdisc *qdisc;
71         int ret;
72
73         qdisc = prio_classify(skb, sch, &ret);
74 #ifdef CONFIG_NET_CLS_ACT
75         if (qdisc == NULL) {
76
77                 if (ret & __NET_XMIT_BYPASS)
78                         sch->qstats.drops++;
79                 kfree_skb(skb);
80                 return ret;
81         }
82 #endif
83
84         ret = qdisc_enqueue(skb, qdisc);
85         if (ret == NET_XMIT_SUCCESS) {
86                 sch->bstats.bytes += qdisc_pkt_len(skb);
87                 sch->bstats.packets++;
88                 sch->q.qlen++;
89                 return NET_XMIT_SUCCESS;
90         }
91         if (net_xmit_drop_count(ret))
92                 sch->qstats.drops++;
93         return ret;
94 }
95
96
97 static int
98 prio_requeue(struct sk_buff *skb, struct Qdisc* sch)
99 {
100         struct Qdisc *qdisc;
101         int ret;
102
103         qdisc = prio_classify(skb, sch, &ret);
104 #ifdef CONFIG_NET_CLS_ACT
105         if (qdisc == NULL) {
106                 if (ret & __NET_XMIT_BYPASS)
107                         sch->qstats.drops++;
108                 kfree_skb(skb);
109                 return ret;
110         }
111 #endif
112
113         if ((ret = qdisc->ops->requeue(skb, qdisc)) == NET_XMIT_SUCCESS) {
114                 sch->q.qlen++;
115                 sch->qstats.requeues++;
116                 return NET_XMIT_SUCCESS;
117         }
118         if (net_xmit_drop_count(ret))
119                 sch->qstats.drops++;
120         return ret;
121 }
122
123
124 static struct sk_buff *prio_dequeue(struct Qdisc* sch)
125 {
126         struct prio_sched_data *q = qdisc_priv(sch);
127         int prio;
128
129         for (prio = 0; prio < q->bands; prio++) {
130                 struct Qdisc *qdisc = q->queues[prio];
131                 struct sk_buff *skb = qdisc->dequeue(qdisc);
132                 if (skb) {
133                         sch->q.qlen--;
134                         return skb;
135                 }
136         }
137         return NULL;
138
139 }
140
141 static unsigned int prio_drop(struct Qdisc* sch)
142 {
143         struct prio_sched_data *q = qdisc_priv(sch);
144         int prio;
145         unsigned int len;
146         struct Qdisc *qdisc;
147
148         for (prio = q->bands-1; prio >= 0; prio--) {
149                 qdisc = q->queues[prio];
150                 if (qdisc->ops->drop && (len = qdisc->ops->drop(qdisc)) != 0) {
151                         sch->q.qlen--;
152                         return len;
153                 }
154         }
155         return 0;
156 }
157
158
159 static void
160 prio_reset(struct Qdisc* sch)
161 {
162         int prio;
163         struct prio_sched_data *q = qdisc_priv(sch);
164
165         for (prio=0; prio<q->bands; prio++)
166                 qdisc_reset(q->queues[prio]);
167         sch->q.qlen = 0;
168 }
169
170 static void
171 prio_destroy(struct Qdisc* sch)
172 {
173         int prio;
174         struct prio_sched_data *q = qdisc_priv(sch);
175
176         tcf_destroy_chain(&q->filter_list);
177         for (prio=0; prio<q->bands; prio++)
178                 qdisc_destroy(q->queues[prio]);
179 }
180
181 static int prio_tune(struct Qdisc *sch, struct nlattr *opt)
182 {
183         struct prio_sched_data *q = qdisc_priv(sch);
184         struct tc_prio_qopt *qopt;
185         int i;
186
187         if (nla_len(opt) < sizeof(*qopt))
188                 return -EINVAL;
189         qopt = nla_data(opt);
190
191         if (qopt->bands > TCQ_PRIO_BANDS || qopt->bands < 2)
192                 return -EINVAL;
193
194         for (i=0; i<=TC_PRIO_MAX; i++) {
195                 if (qopt->priomap[i] >= qopt->bands)
196                         return -EINVAL;
197         }
198
199         sch_tree_lock(sch);
200         q->bands = qopt->bands;
201         memcpy(q->prio2band, qopt->priomap, TC_PRIO_MAX+1);
202
203         for (i=q->bands; i<TCQ_PRIO_BANDS; i++) {
204                 struct Qdisc *child = xchg(&q->queues[i], &noop_qdisc);
205                 if (child != &noop_qdisc) {
206                         qdisc_tree_decrease_qlen(child, child->q.qlen);
207                         qdisc_destroy(child);
208                 }
209         }
210         sch_tree_unlock(sch);
211
212         for (i=0; i<q->bands; i++) {
213                 if (q->queues[i] == &noop_qdisc) {
214                         struct Qdisc *child;
215                         child = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue,
216                                                   &pfifo_qdisc_ops,
217                                                   TC_H_MAKE(sch->handle, i + 1));
218                         if (child) {
219                                 sch_tree_lock(sch);
220                                 child = xchg(&q->queues[i], child);
221
222                                 if (child != &noop_qdisc) {
223                                         qdisc_tree_decrease_qlen(child,
224                                                                  child->q.qlen);
225                                         qdisc_destroy(child);
226                                 }
227                                 sch_tree_unlock(sch);
228                         }
229                 }
230         }
231         return 0;
232 }
233
234 static int prio_init(struct Qdisc *sch, struct nlattr *opt)
235 {
236         struct prio_sched_data *q = qdisc_priv(sch);
237         int i;
238
239         for (i=0; i<TCQ_PRIO_BANDS; i++)
240                 q->queues[i] = &noop_qdisc;
241
242         if (opt == NULL) {
243                 return -EINVAL;
244         } else {
245                 int err;
246
247                 if ((err= prio_tune(sch, opt)) != 0)
248                         return err;
249         }
250         return 0;
251 }
252
253 static int prio_dump(struct Qdisc *sch, struct sk_buff *skb)
254 {
255         struct prio_sched_data *q = qdisc_priv(sch);
256         unsigned char *b = skb_tail_pointer(skb);
257         struct tc_prio_qopt opt;
258
259         opt.bands = q->bands;
260         memcpy(&opt.priomap, q->prio2band, TC_PRIO_MAX+1);
261
262         NLA_PUT(skb, TCA_OPTIONS, sizeof(opt), &opt);
263
264         return skb->len;
265
266 nla_put_failure:
267         nlmsg_trim(skb, b);
268         return -1;
269 }
270
271 static int prio_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new,
272                       struct Qdisc **old)
273 {
274         struct prio_sched_data *q = qdisc_priv(sch);
275         unsigned long band = arg - 1;
276
277         if (band >= q->bands)
278                 return -EINVAL;
279
280         if (new == NULL)
281                 new = &noop_qdisc;
282
283         sch_tree_lock(sch);
284         *old = q->queues[band];
285         q->queues[band] = new;
286         qdisc_tree_decrease_qlen(*old, (*old)->q.qlen);
287         qdisc_reset(*old);
288         sch_tree_unlock(sch);
289
290         return 0;
291 }
292
293 static struct Qdisc *
294 prio_leaf(struct Qdisc *sch, unsigned long arg)
295 {
296         struct prio_sched_data *q = qdisc_priv(sch);
297         unsigned long band = arg - 1;
298
299         if (band >= q->bands)
300                 return NULL;
301
302         return q->queues[band];
303 }
304
305 static unsigned long prio_get(struct Qdisc *sch, u32 classid)
306 {
307         struct prio_sched_data *q = qdisc_priv(sch);
308         unsigned long band = TC_H_MIN(classid);
309
310         if (band - 1 >= q->bands)
311                 return 0;
312         return band;
313 }
314
315 static unsigned long prio_bind(struct Qdisc *sch, unsigned long parent, u32 classid)
316 {
317         return prio_get(sch, classid);
318 }
319
320
321 static void prio_put(struct Qdisc *q, unsigned long cl)
322 {
323         return;
324 }
325
326 static int prio_change(struct Qdisc *sch, u32 handle, u32 parent, struct nlattr **tca, unsigned long *arg)
327 {
328         unsigned long cl = *arg;
329         struct prio_sched_data *q = qdisc_priv(sch);
330
331         if (cl - 1 > q->bands)
332                 return -ENOENT;
333         return 0;
334 }
335
336 static int prio_delete(struct Qdisc *sch, unsigned long cl)
337 {
338         struct prio_sched_data *q = qdisc_priv(sch);
339         if (cl - 1 > q->bands)
340                 return -ENOENT;
341         return 0;
342 }
343
344
345 static int prio_dump_class(struct Qdisc *sch, unsigned long cl, struct sk_buff *skb,
346                            struct tcmsg *tcm)
347 {
348         struct prio_sched_data *q = qdisc_priv(sch);
349
350         if (cl - 1 > q->bands)
351                 return -ENOENT;
352         tcm->tcm_handle |= TC_H_MIN(cl);
353         if (q->queues[cl-1])
354                 tcm->tcm_info = q->queues[cl-1]->handle;
355         return 0;
356 }
357
358 static int prio_dump_class_stats(struct Qdisc *sch, unsigned long cl,
359                                  struct gnet_dump *d)
360 {
361         struct prio_sched_data *q = qdisc_priv(sch);
362         struct Qdisc *cl_q;
363
364         cl_q = q->queues[cl - 1];
365         if (gnet_stats_copy_basic(d, &cl_q->bstats) < 0 ||
366             gnet_stats_copy_queue(d, &cl_q->qstats) < 0)
367                 return -1;
368
369         return 0;
370 }
371
372 static void prio_walk(struct Qdisc *sch, struct qdisc_walker *arg)
373 {
374         struct prio_sched_data *q = qdisc_priv(sch);
375         int prio;
376
377         if (arg->stop)
378                 return;
379
380         for (prio = 0; prio < q->bands; prio++) {
381                 if (arg->count < arg->skip) {
382                         arg->count++;
383                         continue;
384                 }
385                 if (arg->fn(sch, prio+1, arg) < 0) {
386                         arg->stop = 1;
387                         break;
388                 }
389                 arg->count++;
390         }
391 }
392
393 static struct tcf_proto ** prio_find_tcf(struct Qdisc *sch, unsigned long cl)
394 {
395         struct prio_sched_data *q = qdisc_priv(sch);
396
397         if (cl)
398                 return NULL;
399         return &q->filter_list;
400 }
401
402 static const struct Qdisc_class_ops prio_class_ops = {
403         .graft          =       prio_graft,
404         .leaf           =       prio_leaf,
405         .get            =       prio_get,
406         .put            =       prio_put,
407         .change         =       prio_change,
408         .delete         =       prio_delete,
409         .walk           =       prio_walk,
410         .tcf_chain      =       prio_find_tcf,
411         .bind_tcf       =       prio_bind,
412         .unbind_tcf     =       prio_put,
413         .dump           =       prio_dump_class,
414         .dump_stats     =       prio_dump_class_stats,
415 };
416
417 static struct Qdisc_ops prio_qdisc_ops __read_mostly = {
418         .next           =       NULL,
419         .cl_ops         =       &prio_class_ops,
420         .id             =       "prio",
421         .priv_size      =       sizeof(struct prio_sched_data),
422         .enqueue        =       prio_enqueue,
423         .dequeue        =       prio_dequeue,
424         .requeue        =       prio_requeue,
425         .drop           =       prio_drop,
426         .init           =       prio_init,
427         .reset          =       prio_reset,
428         .destroy        =       prio_destroy,
429         .change         =       prio_tune,
430         .dump           =       prio_dump,
431         .owner          =       THIS_MODULE,
432 };
433
434 static int __init prio_module_init(void)
435 {
436         return register_qdisc(&prio_qdisc_ops);
437 }
438
439 static void __exit prio_module_exit(void)
440 {
441         unregister_qdisc(&prio_qdisc_ops);
442 }
443
444 module_init(prio_module_init)
445 module_exit(prio_module_exit)
446
447 MODULE_LICENSE("GPL");